Computer vision systems and methods for detecting and aligning land property boundaries on aerial imagery

Abstract

Systems and methods for detecting and aligning land property boundaries on aerial imagery are provided. The system receives an aerial imagery having land properties. The system applies a feature encoder having a plurality of levels to the aerial imagery. A first level of the plurality of levels includes a convolution block and a discrete wavelet transform layer. The discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band. The high-frequency band is cached and processed with side-convolutional blocks before the high-frequency band are passed to a feature decoder. The system applies the feature decoder to an output of the feature encoder based at least in part on one of inverse discrete wavelet transform layers. The system determines boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function.

The invention claimed is: 
     
       1. A system for detecting land property boundaries on aerial imagery, comprising:
 a memory; and 
 a processor in communication with the memory, the processor: 
 receiving an image having one or more land properties; 
 
       processing the image using a feature encoder having a plurality of levels to the image, a first level of the plurality of levels comprising a convolution block and a discrete wavelet transform layer, wherein the discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band, wherein the high-frequency band is cached and processed with one or more side-convolutional blocks before the high-frequency band is passed to a feature decoder;
 processing an output of the feature encoder based at least in part on a plurality of inverse discrete wavelet transform layers; and 
 determining boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function. 
 
     
     
       2. The system of  claim 1 , wherein the processor processes the image by:
 refining, via the one or more side-convolutional blocks, high-frequency bands collected from each level of the plurality of levels to be coherent with each other; and 
 providing, via the one or more side-convolutional blocks, the refined high-frequency bands to at least one of the plurality of inverse discrete wavelet transform layers. 
 
     
     
       3. The system of  claim 1 , wherein the processor processes the image by:
 providing the high-frequency band to at least one of the plurality of inverse discrete wavelet transform layers. 
 
     
     
       4. The system of  claim 1 , wherein the processor applies a first segmentation block prior to applying at least one inverse discrete wavelet transform layer of the plurality of inverse discrete wavelet transform layers. 
     
     
       5. The system of  claim 4 , wherein the processor applies a second segmentation block subsequent to applying the at least one inverse discrete wavelet transform layer. 
     
     
       6. The system of  claim 1 , wherein determining the boundaries of the one or more land properties is further based at least in part on a non-boundary suppression loss function. 
     
     
       7. The system of  claim 1 , wherein the boundaries of the one or more land properties include road-connected boundaries, and/or boundaries that divide the land property from neighbors. 
     
     
       8. The system of  claim 1 , wherein the processor applies an atrous spatial pyramidal pooling layer to the output of the feature encoder. 
     
     
       9. The system of  claim 1 , wherein the processor further aligns geo-parcel boundaries with the boundaries of the one or more land properties by:
 projecting the geo-parcel boundaries onto a corresponding geo-tagged coordinate system associated with the aerial imagery; 
 determining differences between the geo-parcel boundaries and the boundaries of the one or more land properties; and 
 aligning the geo-parcel boundaries with the boundaries of the one or more land properties based at least in part on the differences. 
 
     
     
       10. A method for detecting land property boundaries on aerial imagery, comprising:
 receiving an image having one or more land properties; 
 processing the image using a feature encoder having a plurality of levels, a first level of the plurality of levels comprising a convolution block and a discrete wavelet transform layer, wherein the discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band, wherein the high-frequency band is cached and processed with one or more side-convolutional blocks before the high-frequency band is passed to a feature decoder; 
 processing an output of the feature encoder based at least in part on a plurality of inverse discrete wavelet transform layers; and 
 determining boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function. 
 
     
     
       11. The method of  claim 10 , wherein processing the image using the feature encoder comprises the steps of:
 refining, via the one or more side-convolutional blocks, high-frequency bands collected from each level of the plurality of levels to be coherent with each other; and 
 providing, via the one or more side-convolutional blocks, the refined high-frequency bands to at least one of the plurality of inverse discrete wavelet transform layers. 
 
     
     
       12. The method of  claim 10 , wherein processing the image comprises the step of providing the high-frequency band to at least one of the plurality of inverse discrete wavelet transform layers. 
     
     
       13. The method of  claim 10 , wherein processing the output of the feature encoder comprises the step of applying a first segmentation block prior to applying at least one inverse discrete wavelet transform layer of the plurality of inverse discrete wavelet transform layers. 
     
     
       14. The method of  claim 13 , further comprising the step of applying a second segmentation block subsequent to applying the at least one inverse discrete wavelet transform layer. 
     
     
       15. The method of  claim 10 , wherein determining the boundaries of the one or more land properties is further based at least in part on a non-boundary suppression loss function. 
     
     
       16. The method of  claim 10 , wherein the boundaries of the one or more land properties include road-connected boundaries, and/or boundaries that divide the land property from neighbors. 
     
     
       17. The method of  claim 10 , further comprising applying an atrous spatial pyramidal pooling layer to the output of the feature encoder. 
     
     
       18. The method of  claim 10 , further comprising:
 projecting geo-parcel boundaries onto a corresponding geo-tagged coordinate system associated with the aerial imagery; 
 determining differences between the geo-parcel boundaries and the boundaries of the one or more land properties; and 
 aligning the geo-parcel boundaries with the boundaries of the one or more land properties based at least in part on the differences. 
 
     
     
       19. A non-transitory computer readable medium having instructions stored thereon for detecting land property boundaries on aerial imagery which, when executed by a processor, causes the processor to carry out the steps of:
 receiving an image having one or more land properties; 
 processing the image using a feature encoder having a plurality of levels, a first level of the plurality of levels comprising a convolution block and a discrete wavelet transform layer, wherein the discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band, wherein the high-frequency band is cached and processed with one or more side-convolutional blocks before the high-frequency band is passed to a feature decoder; 
 processing an output of the feature encoder based at least in part on a plurality of inverse discrete wavelet transform layers; and 
 determining boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function. 
 
     
     
       20. The non-transitory computer readable medium of  claim 19 , wherein processing the image comprises the steps of:
 refining, via the one or more side-convolutional blocks, high-frequency bands collected from each level of the plurality of levels to be coherent with each other; and 
 providing, via the one or more side-convolutional blocks, the refined high-frequency bands to at least one of the plurality of inverse discrete wavelet transform layers. 
 
     
     
       21. The non-transitory computer readable medium of  claim 19 , wherein processing the image comprises the step of providing the high-frequency band to at least one of the plurality of inverse discrete wavelet transform layers. 
     
     
       22. The non-transitory computer readable medium of  claim 19 , wherein processing the output of the feature encoder comprises the step of applying a first segmentation block prior to applying at least one inverse discrete wavelet transform layer of the plurality of inverse discrete wavelet transform layers. 
     
     
       23. The non-transitory computer readable medium of  claim 22 , further comprising the step of applying a second segmentation block subsequent to applying the at least one inverse discrete wavelet transform layer. 
     
     
       24. The non-transitory computer readable medium of  claim 19 , wherein determining the boundaries of the one or more land properties is further based at least in part on a non-boundary suppression loss function. 
     
     
       25. The non-transitory computer readable medium of  claim 19 , wherein the boundaries of the one or more land properties include road-connected boundaries, and/or boundaries that divide the land property from neighbors. 
     
     
       26. The non-transitory computer readable medium of  claim 19 , further comprising applying an atrous spatial pyramidal pooling layer to the output of the feature encoder. 
     
     
       27. The non-transitory computer readable medium of  claim 19 , further comprising:
 projecting geo-parcel boundaries onto a corresponding geo-tagged coordinate system associated with aerial imagery; 
 determining differences between the geo-parcel boundaries and the boundaries of the one or more land properties; and 
 aligning the geo-parcel boundaries with the boundaries of the one or more land properties based at least in part on the differences.